Four MRI-Based Radiomics Models for Diagnosis of Lumbar Intervertebral Disc Degeneration

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Lumbar intervertebral disc degeneration (LIDD) is a primary cause of low back pain.
• Diagnosing LIDD is challenging due to subtle and variable MRI features.
• Accurate early diagnosis is crucial for effective patient management.

Purpose of the Study:

• To develop and validate an MRI-based radiomics ensemble model for precise disc-level LIDD discrimination.
• To address the statistical challenge of non-independent lumbar discs within patients.
• To enhance early and objective LIDD diagnosis in clinical settings.

Main Methods:

• Retrospective analysis of 122 subjects (102 LIDD patients, 20 controls) with 610 lumbar discs.
• Segmentation of intervertebral discs on FS-T2WI MRI and extraction of 1409 IBSI-compliant radiomic features.
• Multi-step feature selection and a soft-voting ensemble model trained with patient-level cross-validation, incorporating GEE and FDR correction.

Main Results:

• The ensemble model achieved excellent diagnostic performance in an independent test set.
• The soft-voting ensemble model demonstrated the highest discrimination with an AUC of 0.976.
• Key intensity and texture radiomic features driving predictions were identified using SHAP analysis.

Conclusions:

• An MRI-based radiomics ensemble model, rigorously corrected for patient-level data, enables accurate and interpretable LIDD discrimination.
• The model achieved high AUC, sensitivity (88%), and specificity (96%) for disc-level LIDD diagnosis.
• This approach holds significant potential for assisting in the early detection and objective diagnosis of LIDD.